| ¡¡ | Chinese Journal of Computers Full Text |
| Title | Internet Traffic Modeling and Prediction Using FARIMA Models |
| Authors | SHU Yan-Tai£±£© WANG Lei£±£© ZHANG Lian-Fang£±£© XUE Fei£±£© JIN Zhi-Gang£±£© Oliver Yang£²£© |
| Address | £±£©(Department of Computer Science, Tianjin University, Tianjin 300072) £²£©(School of Information Technology and Engineering, University of Ottawa, Ottawa, Ontario, Canada K1N 6N5) |
| Year | 2001 |
| Issue | No.1(46-54) |
| Abstract & Background | FARIMA(p,d,q) model is a good traffic model capable of capturing both the long-range and short-range behavior of a network traffic stream in time. In this paper, we provide procedures to model, generate, and predict traffic using FARIMA(p,d,q) model. We suggest a method to simplify the FARIMA model fitting procedure and hence to reduce the time of traffic modeling and prediction. We propose a fitting method to fit a FARIMA model to the actual traffic trace. In essence, this method has the feature of transferring the FARIMA problem into an ARMA problem so that any algorithm (not necessarily related to the method used for fitting the ARMA model) could be used to determine the Hurst parameter. Our method first determines the Hurst parameter (differencing level d) before fitting ARMA model, thus reducing the number of plausible models to be examined and the time to identify model iteratively. Consequently, one can reduce the time to build traffic models and to do real-time modeling. We use the techniques of fractal de-filter (fractional differencing), auxiliary AR model-based backward-prediction, and a combination of rough and accurate parameter estimation in our guidelines to simplify and hence to reduce the time of the model fitting procedure. The accuracy of the method is tested by comparing autocorrelation functions of the original data with those of simulation from the fitted model. We propose an adjusted traffic prediction method using FARIMA models and do many prediction experiments on the actually measured Internet traces. Unlike the previous comments on the complexity of the FARIMA models, we show that FARIMA models could be used for traffic modeling and prediction. Our feasibility-study experiments showed that FARIMA models with less parameters could be used to model and predict actual traffic on quite a large time scale. Our experiments also illustrated that h-step forecasts based on FARIMA model are better than those based on AR model.
keywords long-range dependence, self-similarity, FARIMA, network traffic modeling, network prediction |